fish assemblages - McGARVEY ECOLOGY
Ó 2012 John Wiley & Sons A/S
Ecology of Freshwater Fish 2012: 21: 617–626 Printed in Malaysia All rights reserved
ECOLOGY OF FRESHWATER FISH
Differential saturation of Pacific Northwest and Southeast (USA) fish assemblages Daniel J. McGarvey Center for Environmental Studies, Virginia Commonwealth University, Richmond, VA USA Accepted for publication June 7, 2012
Abstract – Plots of local versus regional richness have been used to test whether assemblages are ‘saturated’ with species. However, the validity of these tests is limited by scale-dependence and arbitrarily defined sampling units, statistical autocorrelation between local and regional richness data, and the confounding effects of propagule pressure. In this study, local versus regional richness plots were used to test the saturation hypothesis for Pacific Northwest and Southeast (USA) fish assemblages, taking care to account for each of the above problems. Specifically, longitudinal river zones were used to ensure that the regional sampling units were not scale-dependent or biologically arbitrary. A log-ratio transformation was used to remove autocorrelation from the local and regional richness data. And a comprehensive fish stocking database was used to account for propagule pressure. Results suggest that the Pacific Northwest fish assemblages, which have low native richness, are not saturated, but the species-rich Southeast assemblages are at or approaching saturation. Key words: local versus regional richness; regional species pool; non-native species; propagule pressure; longitudinal zonation
Introduction
Nonnative fishes are rapidly becoming established in U.S. freshwaters (Mitchell & Knouft 2009). So far, nonnative fish invasions have tended to outpace native extirpations in the United States, and the net effect has been one of homogenisation; overall richness has increased while differences between fish assemblages have decreased (Rahel 2000). But nonnative invasions have not occurred at equal rates throughout the United States. They are most common in the west, where native fish richness is relatively low (Schade & Bonar 2005). In the east, where native fish richness is higher, nonnative invasions are less prevalent (Gido & Brown 1999). Noting that numbers of native and nonnative fish species are often inversely correlated within a given system (see Ross 1991; Gido & Brown 1999), some researchers have suggested that biotic resistance may be a key determinant of invasion success (e.g., Tonn et al. 1990; Angermeier & Winston 1998). The underlying assumption is that all else being equal,
interspecific competition for limiting resources should be more intense in systems with many resident, native species than in systems with few native species (Kennedy et al. 2002). In this way, nonnative species are more likely to survive and become established in species-poor systems. By the same logic, richness may increase over time until a system reaches its carrying capacity and becomes effectively ‘saturated’ with species (Case 1990). Local versus regional richness (LRR) plots are simple tools that have been used to test the saturation hypothesis for a variety of freshwater fish assemblages (e.g., Angermeier & Winston 1998; Oberdorff et al. 1998). LRR plots model average local richness (dependent variable) as a function of the total richness within a larger region (independent variable). By convention, an asymptotic or curvilinear LRR model indicates that local richness is constrained by locally realised processes, such as interspecific competition, and that local assemblages are saturated. Alternatively, a linear LRR model suggests that local communities may be
Correspondence: D. J. McGarvey, Center for Environmental Studies, Virginia Commonwealth University, 1000 West Cary Street, Richmond, VA 23284-3050, USA. E-mail: [email protected]
doi: 10.1111/j.1600-0633.2012.00583.x
617
McGarvey (a)
W
B C
J U
T
N 200 km
(b)
1000 750 80 500
40
0
250
U
W
J
B
C
T
Discharge (m3/s)
Species richness
120
0
Fig. 1. (a) Map of the six study basins, in the states of Oregon (OR) and Alabama (AL). North-west rivers include the Umpqua (U; basin area = 12,133 km2), Willamette (W; area = 29,728 km2) and John Day (J; area = 20,980 km2). South-east rivers include the Black Warrior (B; area = 16,252 km2), Cahaba (C; area = 4730 km2) and Tallapoosa (T; area = 12,108 km2). The OR and AL maps are plotted at a common scale. (b) Fish species richness and discharge in each river basin. Numbers of native and nonnative species in each basin are illustrated by black and white bars, respectively. Grey bars show mean annual discharge near the mouth of each river. Richness data were obtained from Boschung & Mayden (2004), McGarvey & Hughes (2008) and McGarvey and Ward (2008). Discharge data were obtained from the U.S. Geological Survey (http://water.usgs.gov/).
random subsets of a larger pool of potential colonists – the regional species pool (RSP) – and are therefore unsaturated (Cornell 1999). Unfortunately, traditional LRR models are limited by several conceptual and statistical problems, three of which are addressed here. First, LRR results are often scale-dependent and prone to bias when biologically arbitrary definitions of local and/or regional assemblages are used (Srivastava 1999; Loreau 2000). Investigators must therefore use explicit, biologically defensible criteria to delineate local and regional assemblages (see ‘Estimating regional and local richness’ below). Second, standard regression methods should not be used to analyse the LRR relationship because local and regional richness are autocorrleated; by definition, local richness will always be less than or equal to regional richness 618
(Cresswell et al. 1995). A statistical tool to remove this autocorrelation is therefore needed (see SzavaKovats et al. 2012). Third, low numbers of nonnative species may reflect biotic resistance or saturation, but they may also be artefacts of low propagule pressure. On average, nonnative species are more likely to become established in systems where many nonnative individuals are introduced or where repetitive nonnative introductions, such as stocking events, occur (i.e., where propagule pressure is high; Marchetti et al. 2004; Leprieur et al. 2008). Thus, one should not infer that a given assemblage is saturated and therefore resistant to invasion without first verifying that it has been exposed to a moderate to high level of propagule pressure (Gido & Brown 1999). Low propagule pressure weakens support for the saturation hypothesis.
Fish assemblage saturation In this study, I use improved LRR methods to address each of the above problems and test the saturation hypothesis for fish assemblages in Pacific Northwest and Southeast U.S. rivers. I begin by using longitudinal river zones, which are characterised by co-occurring species within relatively homogenous environments, to estimate regional richness (see ‘Estimating regional and local richness’ below). Next, I use the log-ratio transformation of Szava-Kovats et al. (2012) to account for autocorrelation in the local and regional richness data, thereby ensuring that the LRR results are statistically valid. I then use a national fish stocking database (Halverson 2008) to assess whether the LRR results may be artefacts of differential propagule pressure. Northwest and Southeast rivers were chosen because they exhibit a strong gradient in fish diversity, with higher native richness in Southeast rivers, and this gradient reflects the disparate environmental histories of the two biogeographical regions. Southeast U.S. rivers are relatively stable environments with low historical extinction rates, while Northwest U.S. rivers are historically unstable environments where numerous extinctions have occurred (Smith 1981; Moyle & Herbold 1987). I therefore hypothesised that the Southeast fish assemblages are at or near saturation, while the Northwest fish assemblages are not. Material and methods Study sites
Three Pacific Northwest (Umpqua, Willamette and John Day) and three Southeast (Black Warrior, Cahaba and Tallapoosa) rivers were included in this study (Fig. 1a). These rivers were chosen because they reflect the interregional trend in fish richness described above: the Northwest rivers have low native fish richness and high nonnative richness, while the Southeast rivers have high native fish richness and low nonnative richness (Fig. 1b). Notably, this disparity in native richness is not a simple effect of river size. The relatively depauperate Northwest rivers are comparable to or larger than the more diverse Southeast rivers, both in terms of total basin area and river discharge (Fig. 1b). Thus, the interregional differences cannot be attributed to a simple species-area or speciesdischarge effect. Estimating regional and local richness
The first step that must be undertaken in an LRR study is to characterise the RSP in a biologically meaningful way. By definition, the RSP should
include only those species that are physically capable of migrating to and surviving in a typical locality within a given region, or would be in the absence of interspecific competition and predation (Srivastava 1999). Fish studies have often used geographical units such as river basins, physiographical provinces or lakes as RSPs (e.g., Griffiths 1997; Angermeier & Winston 1998). However, these units will overestimate the true size of the RSP if they combine species that have disparate habitat requirements or are otherwise unlikely to co-occur. For instance, a distinct longitudinal gradient in fish species’ distributions and turnover was reported by Li et al. (1987; see their Fig. 24.2) for Pacific Northwest rivers. Strong gradients and discontinuities, such as high species turnover at the fall line, also exist in Southeast rivers (e.g., Mettee et al. 1996). LRR studies that use large river basins as RSPs without accounting for species’ turnover within basins may therefore overestimate the true size of the RSP. Longitudinal zones can potentially solve this problem. When examined at large spatial scales, such as complete river basins, many riverine fish assemblages can be divided into a series of adjacent, zonal subassemblages (Huet 1959; Hawkes 1975). Zonal structure is quantified using multivariate ordinations or clustering algorithms to locate longitudinal discontinuities in assemblage structure (McGarvey & Ward 2008). Adjacent locations that exhibit high turnover in species composition are indicative of zone boundaries, while locations between zone boundaries have comparatively similar assemblage structure. Longitudinal zones typically range from 10 to 100 km in length (e.g., Balon & Stewart 1983) and are often congruent with longitudinal transitions in physical habitat variables, such as water temperature and substrate composition (Hawkes 1975; Statzner & Borchardt 1992; Lamouroux et al. 2002). Thus, longitudinal zones consist of species that co-occur within relatively uniform environments. Similar zonation patterns have been described in each of the six rivers considered here. Using extensive fish assemblage data and common statistical methods, McGarvey & Hughes (2008) and McGarvey & Ward (2008) detected ‘upper’, ‘middle’ and ‘lower’ zones in each of the Pacific Northwest and Southeast rivers. In each river, the succession from upper to middle to lower fish zones coincided with a transition from small, cool, high-gradient environments with few resident species to large, warm, low-gradient habitats with comparatively rich, mixed fish assemblages. Zones were also of similar size among biogeographical regions. For example, mean wetted channel widths were, respectively, 11, 42 and 106 m in the upper, middle and lower zones of the Northwest rivers and 8, 38 and 99 m in the upper, 619
McGarvey middle and lower zones of the Southeast rivers. Upper zone fish assemblages were composed primarily of sculpins (Cottus sp.) and cutthroat trout (Oncorhynchus clarki) in the Northwest rivers and small-bodied minnows (e.g., Cyprinella sp. and Notropis sp.), sunfishes (Lepomis sp.) and darters (Etheostoma sp.) in the Southeast rivers. Lower-zone fishes in the Northwest rivers included large-bodied species such as white sturgeon (Acipenser transmontanus), largescale sucker (Catostomus macrocheilus) and northern pikeminnow (Ptychocheilus oregonensis), in addition to small Cyprinids (e.g., Acrocheilus alutaceus, Richardsonius balteatus and Rhinichthys sp.) and a large number of nonnative sunfishes (Micropterus dolomieu and Lepomis sp.) and catfishes (Ictalurus punctatus and Ameiurus sp.). Characteristic species in the lower zones of the Southeast rivers included bowfin (Amia calva), paddlefish (Polyodon spathula), large suckers (e.g., Carpiodes sp. and Moxostoma sp.), Centrarchids (Micropterus sp., Lepomis sp. and Pomoxis sp.), Esocids (Esox americanus and E. niger), catfishes (Pylodictis olivaris, Ameiurus sp. and Ictalurus sp.), gars (Lepisosteus sp.), temperate basses (Morone sp.), freshwater drum (Aplodinotus grunniens) and Clupeids (Alosa sp. and Dorosoma sp.), in addition to many Cyprinids (e.g., Cyprinella sp. and Notropis sp.) and Percids (Sander vitreus, Ammocrypta sp., Etheostoma sp. and Percina sp.). Complete descriptions of the zonal fish assemblages and all data sources are provided in McGarvey & Hughes (2008) and McGarvey & Ward (2008). In each of the Northwest and Southeast rivers, longitudinal zones satisfied the functional RSP criteria – species with overlapping distributions that could potentially co-occur at a given locality (Srivastava 1999) – better than complete river basins. In other systems, the use of complete river basins as RSPs has been justified. For example, Matthews & Robison (1998) used small, headwater drainage basins within the Ozark and Ouachita highlands (Arkansas) as RSPs, noting that these basins contained high-gradient, clear-water habitats that flowed into and were bounded by low-gradient, turbid rivers; fishes within the highland streams were free to move throughout them but were unlikely to enter turbid rivers further downstream. However, using complete river basins as RSPs in the Northwest and Southeast rivers would have pooled many species with disjunct ranges, thereby inflating the sizes of the RSPs. I therefore used longitudinal zones as RSPs and, in each zone, estimated regional richness as the sum of all species known to occur within the zone. To estimate local richness, I began with the local fish assemblage definition of Matthews (1998, p. 2): ‘fish that occur together in a single place, such that 620
they have at least a reasonable opportunity for daily contact with each other’. By this definition, I assumed that conventional field samples could be used as indicators of local fish assemblage structure and richness. Specifically, I used field samples from the U.S. Environmental Protection Agency’s Environmental Monitoring and Assessment Program and the Alabama Geological Survey to characterise local fish assemblage structure in the Northwest and Southeast rivers, respectively. Fish sampling methods used in each of these programmes were standardised by habitat type: single-pass backpack electrofishing was used in shallow, wadeable habitats and boat electrofishing was used in nonwadeable habitats. Sampling effort was also similar among sites and biogeographical regions, with most sampling events lasting between 45 and 90 min (total electrofishing time) and encompassing ~409 the mean channel width (see McGarvey & Hughes (2008), McGarvey & Ward (2008) and references therein). Additionally, post hoc data analysis tools, such as species accumulation curves, were used to verify that the standardised field samples provided reliable, asymptotic estimates of local richness. For instance, Reynolds et al. (2003) and Paller (1995) showed that standardised field protocols consistently captured 90% of all fish species that were locally present in Northwest and Southeast streams, respectively. For these reasons, I was confident that the standardised field samples could be used to identify fishes that potentially interact on a daily, real-time basis (i.e., to define local assemblages). Local richness was then estimated within each longitudinal zone, or RSP, by calculating the mean richness of 10 randomly selected field samples. This resulted in nine LRR data points (3 zones 9 3 rivers) in each of the two biogeographical regions. Multiple field samples were combined into a single local richness average to avoid pseudoreplication; individual samples collected within a common RSP are spatially correlated and will therefore introduce a source of bias when treated as independent replicates in regression analyses (Srivastava 1999). By averaging field samples, I was able to incorporate some of the among-sample variability in my LRR analyses without introducing spatial bias. Error bars (±1 standard deviation) were also included in the LRR plots as measures of among-sample variability. Log-ratio transformation
The original LRR method used ordinary least squares regression of untransformed data to first test whether a significant linear relationship existed between local and regional richness and then to test whether a quadratic term significantly improved the fit of the model
Fish assemblage saturation Northwest – All species 12
r 2 < 0.01 P = 0.94
–0.4
8
–0.8
SL
ln (SL ÷ [SR - SL])
0.0
–1.2
4
–1.6 –2.0
0 2.0
2.5
3.0
3.5
4.0
0
20
40
60
40
60
100
150
100
150
SR
ln (SR)
Northwest – Native species only 12
r2 = 0.34 P = 0.10
–0.4
8
–0.8
SL
ln (SL ÷ [SR - SL])
0.0
–1.2
4
–1.6 –2.0
0 2.0
2.5
3.0
3.5
4.0
0
20
ln (SR)
SR
Southeast – All species 30 r2 = 0.55 P = 0.02
–0.4
20
–0.8
SL
ln (SL ÷ [SR - SL])
0.0
–1.2
10
–1.6 0
–2.0 3.0
3.5
4.0
4.5
0
5.0
50
SR
ln (SR)
Southeast – Native species only 30 r2 = 0.54 P = 0.03
–0.8
20
SL
ln (SL ÷ [SR - SL])
0.0 –0.4
–1.2
10
–1.6 0
–2.0 3.0
3.5
4.0
4.5
5.0
ln (SR)
0
50
SR
Fig. 2. Local (SL) versus regional richness (SR) plots for Northwest and Southeast rivers. Plots are shown for complete species lists (‘All species’) and for species lists that did not include nonnative fishes (‘Native species only’). Log-ratio-transformed SL and SR data are shown on the left, and the untransformed arithmetic data are shown on the right. Each data point reflects a discrete longitudinal zone or regional species pool. Upper, middle and lower zones are indicated by grey diamonds, white circles, and black squares, respectively. SL estimates are the mean averages of 10 randomly selected field samples within each zone; vertical error bars are ±1 standard deviation in local richness.
(i.e., whether the relationship was asymptotic; Cornell 1999). However, this method was biased because untransformed local and regional richness data are autocorrelated. Local richness is always a subset of, and therefore less than or equal to, regional richness. This autocorrelation ensures that the resulting LRR relationship will be constrained to a ‘wedge-shaped’
area in arithmetic space (i.e., the area below the 1:1 line in LRR plots), thereby invalidating ordinary regression results (Cresswell et al. 1995). Szava-Kovats et al. (2012) solved this problem with a log-ratio transformation. They first partitioned regional richness into two additive components: a-diversity (local richness) and b-diversity (regional – 621
McGarvey local richness). Next, they used an additive log-ratio transformation to generate the unconstrained, statistically independent function y = ln(a-diversity/b-diversity). Finally, they plotted y against the natural log of regional richness and used least squares regression to model the relationship. This allowed them to determine whether local richness was a relative, rather than absolute, function of regional richness. Significantly negative slopes provided unbiased evidence of asymptotic LRR relationships (in arithmetic space) while nonsignificant or zero-slope models were indicative of linear, unsaturated LRR relationships (see Fig. 1 in Szava-Kovats et al. 2012). I used this method to perform unbiased LRR tests for the Pacific Northwest and Southeast fish data.
agencies in 2004, including total numbers of individuals and total biomasses stocked, and summarised these data by species and region (western, southern, north central, and north-eastern states). By comparing the numbers and biomasses of fishes stocked in western and southern states, I tested whether propagule pressure differs between the two biogeographical regions. These results were then combined with the LRR test results to assess the overall strength of evidence for the saturation hypothesis. If the LRR test results indicate saturation and the stocking data indicate high propagule pressure, support for the saturation hypothesis is increased. However, if the LRR results reflect saturation but the stocking data do not indicate high propagule pressure, support for the saturation hypothesis is weakened.
Propagule pressure
Large numbers of nonnative species may create the impression that abundant, unoccupied niche space is available within a given system and that native assemblages are unsaturated. Alternatively, low numbers of nonnative species suggests that little unoccupied niche space is available and that native assemblages are saturated. But a valid test of the saturation hypothesis cannot be performed without first accounting for propagule pressure; one cannot prove that an assemblage is invasion resistant without first documenting that invasions have in fact been attempted (see Introduction; Gido & Brown 1999). I used the fish stocking data of Halverson (2008) as an index of propagule pressure in Pacific Northwest and Southeast rivers. Halverson (2008) compiled all stocking efforts by state and federal Table 1. Total numbers (91000 individuals) and biomasses (91000 kg) of fishes stocked in Western and Southern U.S. states in 2004.
All species Non-native species within study rivers Without walleye
Western states†
Southern states‡
Number
Biomass
Number
Biomass
424,820 85,582
10,805 310
111,767 13,344
2,684 1,742
6576
288
–
–
‘All species’ data are inclusive of all stocked fishes. ‘Non-native species within study rivers data are specific to non-native fishes that are stocked in the Pacific Northwest study basins (Umpqua, Willamette John Day) or the Southeast study basins (Black Warrior, Cahaba, Tallapoosa). Adjusted ‘without walleye’ numbers are not shown for the Southern states because walleye are native to the three Southeast rivers. All stocking data are from Halverson (2008). † Western states include Oregon, Washington, Alaska, California, Nevada, Arizona, New Mexico, Utah, Colorado, Wyoming, Idaho, Montana, and Hawaii. ‡ Southern states include Alabama, Mississippi, Louisiana, Texas, Oklahoma, Arkansas, Florida, Georgia, South Carolina, North Carolina, Tennessee, Virginia, West Virginia, Kentucky, and Maryland.
622
Results
When the log-ratio transformed LRR data were plotted for Pacific Northwest fishes, I observed little or no evidence of a significant relationship, negative or otherwise (P 0.10; Fig. 2 upper panel). This was the case when all species were included in the LRR analyses and when only native species were included. Accordingly, the untransformed LRR plots (i.e., local and regional richness plotted in arithmetic space) were linear, rather than asymptotic, and provided no evidence for the saturation hypothesis. However, the LRR analyses did suggest that the Southeast fish assemblages may be at or approaching saturation. Both the all-species data set and the native-speciesonly data set exhibited significant, negative relationships when the log-ratio transformation was applied to the LRR data (Fig. 2 lower panel). The corresponding arithmetic LRR plots therefore exhibited significant curvature for Southeast fishes. When all species were considered, fish stocking effort in the western United States exceeded that in the southern United States by an approximately 4:1 margin, both in terms of total numbers of individuals and total biomass (Table 1). However, these figures were potentially misleading because many of these fishes were native to the systems they were stocked in and should not be included in indices of nonnative propagule pressure. For example, ~50% of the fishes stocked in the western United States were native salmon, and ~70% of the fishes stocked in the southern United States were native sunfishes (Halverson 2008). When the stocking effort data were limited to nonnative species that are known to occur in at least one of the six study rivers (Fig. 1a), the number of individuals stocked in the western United States still exceeded the number stocked in the southern United States, but the total biomass of stocked fishes was greater in the southern United States (Table 1).
Fish assemblage saturation Furthermore, walleye (S. vitreus) comprised >90% of the remaining stocking effort in the western United States. But of the three Pacific Northwest rivers included in this study, walleye are only known to occur in the lower zone of the Willamette River, and they are generally rare within this zone (see McGarvey & Hughes 2008). When walleye were removed from the western stocking data, both the numbers and biomasses of stocked nonnative fishes were greater in the southern United States than in the western Unites States. Thus, the data of Halverson (2008) suggest that nonnative propagule pressure is not greater in Pacific Northwest rivers than in Southeast rivers. This in turn supports the hypothesis that Northwest rivers, which contain many nonnative fishes, are not saturated while Southeast rivers, which contain relatively few nonnative fishes, are at or near saturation. Discussion
This study is, to my knowledge, the first to use LRR plots to test for differential saturation of riverine fish assemblages in two distinct biogeographical regions (but see Tonn et al. 1990 and Griffiths 1997 for lacustrine comparisons). LRR plots revealed no evidence of saturation within Pacific Northwest fish assemblages, but did support the saturation hypothesis for Southeast fish assemblages (Fig. 2). Moreover, these results were robust to methodological problems associated with previous LRR studies. The regional richness estimates, which were derived from longitudinal river zones, and the local richness estimates, which were derived from standardised field samples, were based on explicit biological criteria and did not require the use of arbitrary spatial units (Srivastava 1999; Loreau 2000). The log-ratio transformation of Szava-Kovats et al. (2012) corrected for statistical autocorrelation between local and regional richness (Cresswell et al. 1995). And the fish stocking data of Halverson (2008) supported the LRR results by indicating that nonnative propagule pressure is at least as strong in the Southeast rivers as the Northwest rivers (Table 1). The differential saturation results are intuitive when placed in a historical context. Since the Pleistocene (
Ecology of Freshwater Fish 2012: 21: 617–626 Printed in Malaysia All rights reserved
ECOLOGY OF FRESHWATER FISH
Differential saturation of Pacific Northwest and Southeast (USA) fish assemblages Daniel J. McGarvey Center for Environmental Studies, Virginia Commonwealth University, Richmond, VA USA Accepted for publication June 7, 2012
Abstract – Plots of local versus regional richness have been used to test whether assemblages are ‘saturated’ with species. However, the validity of these tests is limited by scale-dependence and arbitrarily defined sampling units, statistical autocorrelation between local and regional richness data, and the confounding effects of propagule pressure. In this study, local versus regional richness plots were used to test the saturation hypothesis for Pacific Northwest and Southeast (USA) fish assemblages, taking care to account for each of the above problems. Specifically, longitudinal river zones were used to ensure that the regional sampling units were not scale-dependent or biologically arbitrary. A log-ratio transformation was used to remove autocorrelation from the local and regional richness data. And a comprehensive fish stocking database was used to account for propagule pressure. Results suggest that the Pacific Northwest fish assemblages, which have low native richness, are not saturated, but the species-rich Southeast assemblages are at or approaching saturation. Key words: local versus regional richness; regional species pool; non-native species; propagule pressure; longitudinal zonation
Introduction
Nonnative fishes are rapidly becoming established in U.S. freshwaters (Mitchell & Knouft 2009). So far, nonnative fish invasions have tended to outpace native extirpations in the United States, and the net effect has been one of homogenisation; overall richness has increased while differences between fish assemblages have decreased (Rahel 2000). But nonnative invasions have not occurred at equal rates throughout the United States. They are most common in the west, where native fish richness is relatively low (Schade & Bonar 2005). In the east, where native fish richness is higher, nonnative invasions are less prevalent (Gido & Brown 1999). Noting that numbers of native and nonnative fish species are often inversely correlated within a given system (see Ross 1991; Gido & Brown 1999), some researchers have suggested that biotic resistance may be a key determinant of invasion success (e.g., Tonn et al. 1990; Angermeier & Winston 1998). The underlying assumption is that all else being equal,
interspecific competition for limiting resources should be more intense in systems with many resident, native species than in systems with few native species (Kennedy et al. 2002). In this way, nonnative species are more likely to survive and become established in species-poor systems. By the same logic, richness may increase over time until a system reaches its carrying capacity and becomes effectively ‘saturated’ with species (Case 1990). Local versus regional richness (LRR) plots are simple tools that have been used to test the saturation hypothesis for a variety of freshwater fish assemblages (e.g., Angermeier & Winston 1998; Oberdorff et al. 1998). LRR plots model average local richness (dependent variable) as a function of the total richness within a larger region (independent variable). By convention, an asymptotic or curvilinear LRR model indicates that local richness is constrained by locally realised processes, such as interspecific competition, and that local assemblages are saturated. Alternatively, a linear LRR model suggests that local communities may be
Correspondence: D. J. McGarvey, Center for Environmental Studies, Virginia Commonwealth University, 1000 West Cary Street, Richmond, VA 23284-3050, USA. E-mail: [email protected]
doi: 10.1111/j.1600-0633.2012.00583.x
617
McGarvey (a)
W
B C
J U
T
N 200 km
(b)
1000 750 80 500
40
0
250
U
W
J
B
C
T
Discharge (m3/s)
Species richness
120
0
Fig. 1. (a) Map of the six study basins, in the states of Oregon (OR) and Alabama (AL). North-west rivers include the Umpqua (U; basin area = 12,133 km2), Willamette (W; area = 29,728 km2) and John Day (J; area = 20,980 km2). South-east rivers include the Black Warrior (B; area = 16,252 km2), Cahaba (C; area = 4730 km2) and Tallapoosa (T; area = 12,108 km2). The OR and AL maps are plotted at a common scale. (b) Fish species richness and discharge in each river basin. Numbers of native and nonnative species in each basin are illustrated by black and white bars, respectively. Grey bars show mean annual discharge near the mouth of each river. Richness data were obtained from Boschung & Mayden (2004), McGarvey & Hughes (2008) and McGarvey and Ward (2008). Discharge data were obtained from the U.S. Geological Survey (http://water.usgs.gov/).
random subsets of a larger pool of potential colonists – the regional species pool (RSP) – and are therefore unsaturated (Cornell 1999). Unfortunately, traditional LRR models are limited by several conceptual and statistical problems, three of which are addressed here. First, LRR results are often scale-dependent and prone to bias when biologically arbitrary definitions of local and/or regional assemblages are used (Srivastava 1999; Loreau 2000). Investigators must therefore use explicit, biologically defensible criteria to delineate local and regional assemblages (see ‘Estimating regional and local richness’ below). Second, standard regression methods should not be used to analyse the LRR relationship because local and regional richness are autocorrleated; by definition, local richness will always be less than or equal to regional richness 618
(Cresswell et al. 1995). A statistical tool to remove this autocorrelation is therefore needed (see SzavaKovats et al. 2012). Third, low numbers of nonnative species may reflect biotic resistance or saturation, but they may also be artefacts of low propagule pressure. On average, nonnative species are more likely to become established in systems where many nonnative individuals are introduced or where repetitive nonnative introductions, such as stocking events, occur (i.e., where propagule pressure is high; Marchetti et al. 2004; Leprieur et al. 2008). Thus, one should not infer that a given assemblage is saturated and therefore resistant to invasion without first verifying that it has been exposed to a moderate to high level of propagule pressure (Gido & Brown 1999). Low propagule pressure weakens support for the saturation hypothesis.
Fish assemblage saturation In this study, I use improved LRR methods to address each of the above problems and test the saturation hypothesis for fish assemblages in Pacific Northwest and Southeast U.S. rivers. I begin by using longitudinal river zones, which are characterised by co-occurring species within relatively homogenous environments, to estimate regional richness (see ‘Estimating regional and local richness’ below). Next, I use the log-ratio transformation of Szava-Kovats et al. (2012) to account for autocorrelation in the local and regional richness data, thereby ensuring that the LRR results are statistically valid. I then use a national fish stocking database (Halverson 2008) to assess whether the LRR results may be artefacts of differential propagule pressure. Northwest and Southeast rivers were chosen because they exhibit a strong gradient in fish diversity, with higher native richness in Southeast rivers, and this gradient reflects the disparate environmental histories of the two biogeographical regions. Southeast U.S. rivers are relatively stable environments with low historical extinction rates, while Northwest U.S. rivers are historically unstable environments where numerous extinctions have occurred (Smith 1981; Moyle & Herbold 1987). I therefore hypothesised that the Southeast fish assemblages are at or near saturation, while the Northwest fish assemblages are not. Material and methods Study sites
Three Pacific Northwest (Umpqua, Willamette and John Day) and three Southeast (Black Warrior, Cahaba and Tallapoosa) rivers were included in this study (Fig. 1a). These rivers were chosen because they reflect the interregional trend in fish richness described above: the Northwest rivers have low native fish richness and high nonnative richness, while the Southeast rivers have high native fish richness and low nonnative richness (Fig. 1b). Notably, this disparity in native richness is not a simple effect of river size. The relatively depauperate Northwest rivers are comparable to or larger than the more diverse Southeast rivers, both in terms of total basin area and river discharge (Fig. 1b). Thus, the interregional differences cannot be attributed to a simple species-area or speciesdischarge effect. Estimating regional and local richness
The first step that must be undertaken in an LRR study is to characterise the RSP in a biologically meaningful way. By definition, the RSP should
include only those species that are physically capable of migrating to and surviving in a typical locality within a given region, or would be in the absence of interspecific competition and predation (Srivastava 1999). Fish studies have often used geographical units such as river basins, physiographical provinces or lakes as RSPs (e.g., Griffiths 1997; Angermeier & Winston 1998). However, these units will overestimate the true size of the RSP if they combine species that have disparate habitat requirements or are otherwise unlikely to co-occur. For instance, a distinct longitudinal gradient in fish species’ distributions and turnover was reported by Li et al. (1987; see their Fig. 24.2) for Pacific Northwest rivers. Strong gradients and discontinuities, such as high species turnover at the fall line, also exist in Southeast rivers (e.g., Mettee et al. 1996). LRR studies that use large river basins as RSPs without accounting for species’ turnover within basins may therefore overestimate the true size of the RSP. Longitudinal zones can potentially solve this problem. When examined at large spatial scales, such as complete river basins, many riverine fish assemblages can be divided into a series of adjacent, zonal subassemblages (Huet 1959; Hawkes 1975). Zonal structure is quantified using multivariate ordinations or clustering algorithms to locate longitudinal discontinuities in assemblage structure (McGarvey & Ward 2008). Adjacent locations that exhibit high turnover in species composition are indicative of zone boundaries, while locations between zone boundaries have comparatively similar assemblage structure. Longitudinal zones typically range from 10 to 100 km in length (e.g., Balon & Stewart 1983) and are often congruent with longitudinal transitions in physical habitat variables, such as water temperature and substrate composition (Hawkes 1975; Statzner & Borchardt 1992; Lamouroux et al. 2002). Thus, longitudinal zones consist of species that co-occur within relatively uniform environments. Similar zonation patterns have been described in each of the six rivers considered here. Using extensive fish assemblage data and common statistical methods, McGarvey & Hughes (2008) and McGarvey & Ward (2008) detected ‘upper’, ‘middle’ and ‘lower’ zones in each of the Pacific Northwest and Southeast rivers. In each river, the succession from upper to middle to lower fish zones coincided with a transition from small, cool, high-gradient environments with few resident species to large, warm, low-gradient habitats with comparatively rich, mixed fish assemblages. Zones were also of similar size among biogeographical regions. For example, mean wetted channel widths were, respectively, 11, 42 and 106 m in the upper, middle and lower zones of the Northwest rivers and 8, 38 and 99 m in the upper, 619
McGarvey middle and lower zones of the Southeast rivers. Upper zone fish assemblages were composed primarily of sculpins (Cottus sp.) and cutthroat trout (Oncorhynchus clarki) in the Northwest rivers and small-bodied minnows (e.g., Cyprinella sp. and Notropis sp.), sunfishes (Lepomis sp.) and darters (Etheostoma sp.) in the Southeast rivers. Lower-zone fishes in the Northwest rivers included large-bodied species such as white sturgeon (Acipenser transmontanus), largescale sucker (Catostomus macrocheilus) and northern pikeminnow (Ptychocheilus oregonensis), in addition to small Cyprinids (e.g., Acrocheilus alutaceus, Richardsonius balteatus and Rhinichthys sp.) and a large number of nonnative sunfishes (Micropterus dolomieu and Lepomis sp.) and catfishes (Ictalurus punctatus and Ameiurus sp.). Characteristic species in the lower zones of the Southeast rivers included bowfin (Amia calva), paddlefish (Polyodon spathula), large suckers (e.g., Carpiodes sp. and Moxostoma sp.), Centrarchids (Micropterus sp., Lepomis sp. and Pomoxis sp.), Esocids (Esox americanus and E. niger), catfishes (Pylodictis olivaris, Ameiurus sp. and Ictalurus sp.), gars (Lepisosteus sp.), temperate basses (Morone sp.), freshwater drum (Aplodinotus grunniens) and Clupeids (Alosa sp. and Dorosoma sp.), in addition to many Cyprinids (e.g., Cyprinella sp. and Notropis sp.) and Percids (Sander vitreus, Ammocrypta sp., Etheostoma sp. and Percina sp.). Complete descriptions of the zonal fish assemblages and all data sources are provided in McGarvey & Hughes (2008) and McGarvey & Ward (2008). In each of the Northwest and Southeast rivers, longitudinal zones satisfied the functional RSP criteria – species with overlapping distributions that could potentially co-occur at a given locality (Srivastava 1999) – better than complete river basins. In other systems, the use of complete river basins as RSPs has been justified. For example, Matthews & Robison (1998) used small, headwater drainage basins within the Ozark and Ouachita highlands (Arkansas) as RSPs, noting that these basins contained high-gradient, clear-water habitats that flowed into and were bounded by low-gradient, turbid rivers; fishes within the highland streams were free to move throughout them but were unlikely to enter turbid rivers further downstream. However, using complete river basins as RSPs in the Northwest and Southeast rivers would have pooled many species with disjunct ranges, thereby inflating the sizes of the RSPs. I therefore used longitudinal zones as RSPs and, in each zone, estimated regional richness as the sum of all species known to occur within the zone. To estimate local richness, I began with the local fish assemblage definition of Matthews (1998, p. 2): ‘fish that occur together in a single place, such that 620
they have at least a reasonable opportunity for daily contact with each other’. By this definition, I assumed that conventional field samples could be used as indicators of local fish assemblage structure and richness. Specifically, I used field samples from the U.S. Environmental Protection Agency’s Environmental Monitoring and Assessment Program and the Alabama Geological Survey to characterise local fish assemblage structure in the Northwest and Southeast rivers, respectively. Fish sampling methods used in each of these programmes were standardised by habitat type: single-pass backpack electrofishing was used in shallow, wadeable habitats and boat electrofishing was used in nonwadeable habitats. Sampling effort was also similar among sites and biogeographical regions, with most sampling events lasting between 45 and 90 min (total electrofishing time) and encompassing ~409 the mean channel width (see McGarvey & Hughes (2008), McGarvey & Ward (2008) and references therein). Additionally, post hoc data analysis tools, such as species accumulation curves, were used to verify that the standardised field samples provided reliable, asymptotic estimates of local richness. For instance, Reynolds et al. (2003) and Paller (1995) showed that standardised field protocols consistently captured 90% of all fish species that were locally present in Northwest and Southeast streams, respectively. For these reasons, I was confident that the standardised field samples could be used to identify fishes that potentially interact on a daily, real-time basis (i.e., to define local assemblages). Local richness was then estimated within each longitudinal zone, or RSP, by calculating the mean richness of 10 randomly selected field samples. This resulted in nine LRR data points (3 zones 9 3 rivers) in each of the two biogeographical regions. Multiple field samples were combined into a single local richness average to avoid pseudoreplication; individual samples collected within a common RSP are spatially correlated and will therefore introduce a source of bias when treated as independent replicates in regression analyses (Srivastava 1999). By averaging field samples, I was able to incorporate some of the among-sample variability in my LRR analyses without introducing spatial bias. Error bars (±1 standard deviation) were also included in the LRR plots as measures of among-sample variability. Log-ratio transformation
The original LRR method used ordinary least squares regression of untransformed data to first test whether a significant linear relationship existed between local and regional richness and then to test whether a quadratic term significantly improved the fit of the model
Fish assemblage saturation Northwest – All species 12
r 2 < 0.01 P = 0.94
–0.4
8
–0.8
SL
ln (SL ÷ [SR - SL])
0.0
–1.2
4
–1.6 –2.0
0 2.0
2.5
3.0
3.5
4.0
0
20
40
60
40
60
100
150
100
150
SR
ln (SR)
Northwest – Native species only 12
r2 = 0.34 P = 0.10
–0.4
8
–0.8
SL
ln (SL ÷ [SR - SL])
0.0
–1.2
4
–1.6 –2.0
0 2.0
2.5
3.0
3.5
4.0
0
20
ln (SR)
SR
Southeast – All species 30 r2 = 0.55 P = 0.02
–0.4
20
–0.8
SL
ln (SL ÷ [SR - SL])
0.0
–1.2
10
–1.6 0
–2.0 3.0
3.5
4.0
4.5
0
5.0
50
SR
ln (SR)
Southeast – Native species only 30 r2 = 0.54 P = 0.03
–0.8
20
SL
ln (SL ÷ [SR - SL])
0.0 –0.4
–1.2
10
–1.6 0
–2.0 3.0
3.5
4.0
4.5
5.0
ln (SR)
0
50
SR
Fig. 2. Local (SL) versus regional richness (SR) plots for Northwest and Southeast rivers. Plots are shown for complete species lists (‘All species’) and for species lists that did not include nonnative fishes (‘Native species only’). Log-ratio-transformed SL and SR data are shown on the left, and the untransformed arithmetic data are shown on the right. Each data point reflects a discrete longitudinal zone or regional species pool. Upper, middle and lower zones are indicated by grey diamonds, white circles, and black squares, respectively. SL estimates are the mean averages of 10 randomly selected field samples within each zone; vertical error bars are ±1 standard deviation in local richness.
(i.e., whether the relationship was asymptotic; Cornell 1999). However, this method was biased because untransformed local and regional richness data are autocorrelated. Local richness is always a subset of, and therefore less than or equal to, regional richness. This autocorrelation ensures that the resulting LRR relationship will be constrained to a ‘wedge-shaped’
area in arithmetic space (i.e., the area below the 1:1 line in LRR plots), thereby invalidating ordinary regression results (Cresswell et al. 1995). Szava-Kovats et al. (2012) solved this problem with a log-ratio transformation. They first partitioned regional richness into two additive components: a-diversity (local richness) and b-diversity (regional – 621
McGarvey local richness). Next, they used an additive log-ratio transformation to generate the unconstrained, statistically independent function y = ln(a-diversity/b-diversity). Finally, they plotted y against the natural log of regional richness and used least squares regression to model the relationship. This allowed them to determine whether local richness was a relative, rather than absolute, function of regional richness. Significantly negative slopes provided unbiased evidence of asymptotic LRR relationships (in arithmetic space) while nonsignificant or zero-slope models were indicative of linear, unsaturated LRR relationships (see Fig. 1 in Szava-Kovats et al. 2012). I used this method to perform unbiased LRR tests for the Pacific Northwest and Southeast fish data.
agencies in 2004, including total numbers of individuals and total biomasses stocked, and summarised these data by species and region (western, southern, north central, and north-eastern states). By comparing the numbers and biomasses of fishes stocked in western and southern states, I tested whether propagule pressure differs between the two biogeographical regions. These results were then combined with the LRR test results to assess the overall strength of evidence for the saturation hypothesis. If the LRR test results indicate saturation and the stocking data indicate high propagule pressure, support for the saturation hypothesis is increased. However, if the LRR results reflect saturation but the stocking data do not indicate high propagule pressure, support for the saturation hypothesis is weakened.
Propagule pressure
Large numbers of nonnative species may create the impression that abundant, unoccupied niche space is available within a given system and that native assemblages are unsaturated. Alternatively, low numbers of nonnative species suggests that little unoccupied niche space is available and that native assemblages are saturated. But a valid test of the saturation hypothesis cannot be performed without first accounting for propagule pressure; one cannot prove that an assemblage is invasion resistant without first documenting that invasions have in fact been attempted (see Introduction; Gido & Brown 1999). I used the fish stocking data of Halverson (2008) as an index of propagule pressure in Pacific Northwest and Southeast rivers. Halverson (2008) compiled all stocking efforts by state and federal Table 1. Total numbers (91000 individuals) and biomasses (91000 kg) of fishes stocked in Western and Southern U.S. states in 2004.
All species Non-native species within study rivers Without walleye
Western states†
Southern states‡
Number
Biomass
Number
Biomass
424,820 85,582
10,805 310
111,767 13,344
2,684 1,742
6576
288
–
–
‘All species’ data are inclusive of all stocked fishes. ‘Non-native species within study rivers data are specific to non-native fishes that are stocked in the Pacific Northwest study basins (Umpqua, Willamette John Day) or the Southeast study basins (Black Warrior, Cahaba, Tallapoosa). Adjusted ‘without walleye’ numbers are not shown for the Southern states because walleye are native to the three Southeast rivers. All stocking data are from Halverson (2008). † Western states include Oregon, Washington, Alaska, California, Nevada, Arizona, New Mexico, Utah, Colorado, Wyoming, Idaho, Montana, and Hawaii. ‡ Southern states include Alabama, Mississippi, Louisiana, Texas, Oklahoma, Arkansas, Florida, Georgia, South Carolina, North Carolina, Tennessee, Virginia, West Virginia, Kentucky, and Maryland.
622
Results
When the log-ratio transformed LRR data were plotted for Pacific Northwest fishes, I observed little or no evidence of a significant relationship, negative or otherwise (P 0.10; Fig. 2 upper panel). This was the case when all species were included in the LRR analyses and when only native species were included. Accordingly, the untransformed LRR plots (i.e., local and regional richness plotted in arithmetic space) were linear, rather than asymptotic, and provided no evidence for the saturation hypothesis. However, the LRR analyses did suggest that the Southeast fish assemblages may be at or approaching saturation. Both the all-species data set and the native-speciesonly data set exhibited significant, negative relationships when the log-ratio transformation was applied to the LRR data (Fig. 2 lower panel). The corresponding arithmetic LRR plots therefore exhibited significant curvature for Southeast fishes. When all species were considered, fish stocking effort in the western United States exceeded that in the southern United States by an approximately 4:1 margin, both in terms of total numbers of individuals and total biomass (Table 1). However, these figures were potentially misleading because many of these fishes were native to the systems they were stocked in and should not be included in indices of nonnative propagule pressure. For example, ~50% of the fishes stocked in the western United States were native salmon, and ~70% of the fishes stocked in the southern United States were native sunfishes (Halverson 2008). When the stocking effort data were limited to nonnative species that are known to occur in at least one of the six study rivers (Fig. 1a), the number of individuals stocked in the western United States still exceeded the number stocked in the southern United States, but the total biomass of stocked fishes was greater in the southern United States (Table 1).
Fish assemblage saturation Furthermore, walleye (S. vitreus) comprised >90% of the remaining stocking effort in the western United States. But of the three Pacific Northwest rivers included in this study, walleye are only known to occur in the lower zone of the Willamette River, and they are generally rare within this zone (see McGarvey & Hughes 2008). When walleye were removed from the western stocking data, both the numbers and biomasses of stocked nonnative fishes were greater in the southern United States than in the western Unites States. Thus, the data of Halverson (2008) suggest that nonnative propagule pressure is not greater in Pacific Northwest rivers than in Southeast rivers. This in turn supports the hypothesis that Northwest rivers, which contain many nonnative fishes, are not saturated while Southeast rivers, which contain relatively few nonnative fishes, are at or near saturation. Discussion
This study is, to my knowledge, the first to use LRR plots to test for differential saturation of riverine fish assemblages in two distinct biogeographical regions (but see Tonn et al. 1990 and Griffiths 1997 for lacustrine comparisons). LRR plots revealed no evidence of saturation within Pacific Northwest fish assemblages, but did support the saturation hypothesis for Southeast fish assemblages (Fig. 2). Moreover, these results were robust to methodological problems associated with previous LRR studies. The regional richness estimates, which were derived from longitudinal river zones, and the local richness estimates, which were derived from standardised field samples, were based on explicit biological criteria and did not require the use of arbitrary spatial units (Srivastava 1999; Loreau 2000). The log-ratio transformation of Szava-Kovats et al. (2012) corrected for statistical autocorrelation between local and regional richness (Cresswell et al. 1995). And the fish stocking data of Halverson (2008) supported the LRR results by indicating that nonnative propagule pressure is at least as strong in the Southeast rivers as the Northwest rivers (Table 1). The differential saturation results are intuitive when placed in a historical context. Since the Pleistocene (
